What we build

Each of these can be a standalone project or a feature inside something you already run.

Customer support chatbot

Answers questions using your help centre, policies, product data and past tickets. It shows where each answer came from, and hands the conversation to your team when it isn't sure or the customer asks for a person.

Website widget · WhatsApp · Shopify · Zendesk / Freshdesk / Gorgias handover

Shopify store assistant

Helps shoppers find the right product, compare options, check sizing and look up order status. It reads live product and inventory data, so it never recommends something that's out of stock.

Theme app extension · Storefront API · Admin API

AI features inside your product

Search that understands what users mean, summaries, drafting, classification and smart form filling, added to your web app through a clean API layer, with usage limits and cost tracking.

Node.js / Java backends · React / Next.js frontends

Document and data extraction

Reads invoices, purchase orders, forms, CVs or emails and turns them into structured records in your database, ERP or spreadsheet. Anything it isn't confident about goes to a review queue for a person to check.

PDF · scanned images · email inboxes · Google Sheets

Internal knowledge assistant

One place for your team to ask questions across wikis, Google Drive, Notion, Confluence or Slack, respecting who is allowed to see what.

Retrieval (RAG) · access controls · Slack / Teams bots

Automations with an AI step

Existing workflows with a model doing one well-defined job in the middle, such as tagging incoming orders, routing support tickets or drafting replies for approval.

Webhooks · queues · Shopify Flow · n8n / Zapier / Make

Reliability

How we keep it from making things up

  • Answers grounded in your content, with sources shown where it helps.
  • A test set before launch: real questions from your customers, checked against expected answers, and re-run whenever anything changes.
  • Clear limits on what the bot will and won't discuss, and a handover to a person.
  • Logs you can read, so you can see what people asked and how it replied.
  • Cost controls: per-user limits, caching and right-sized models.
Data

Your data stays yours

We use the providers' business APIs, which by default don't use your data to train their models. API keys, logs and vector stores live in your cloud accounts, not ours.

If data can't leave your servers, we can run open-weight models on your own infrastructure instead. They're less capable than the top hosted models, and we'll explain the trade-off for your use case.

Want an assistant like Claude or ChatGPT to use your systems rather than chat on your website? See MCP servers.

Questions

Do you build custom AI chatbots, or set up an off-the-shelf one?

Both. If an existing product, such as your helpdesk's built-in AI, covers what you need, we'll set that up and connect it properly. We build custom when you need your own data, your own rules or your own interface.

Which AI model do you use?

Whichever fits the job and the budget. That's usually Claude or an OpenAI model for customer-facing work, smaller models for high-volume classification, and open-weight models when data must stay on your servers. We build so the model can be swapped later.

How long does a support chatbot take?

We'd expect a first version on your real content to take a few weeks. It depends mostly on how organised your existing content is and which systems it needs to read from. You'll get a precise timeline in the written scope.

What does it cost to run?

Model usage is billed by the provider directly to you. We estimate it from your expected traffic before we build, and set limits so a busy day can't produce a surprise bill.

Have a process that AI could take off your hands?

Describe it in a few lines. We'll tell you honestly whether AI is the right tool for it.

Start a project